For basic connectivity, before considering routing protocols, both options can provide the same connectivity.
Some of the other factors that may go into the design would include which routing protocols are being used, and network types that are allowed or not allowed to be configured.
With the sub-interface option, split horizon issues could be avoided, however, if the main interface is used, EIGRP split horizon could be disabled, which would solve that problem (if EIGRP was being used).
Hope that helps,
Recommended or suggested way of setting up FR is using p-t-p subinterfaces whenever you can;
for those who are getting multiple DLCIs, you must assign one DLCI per PVC,
either frame-relay interface-dlci or frame map ip will work on sub-interfaces; it depends on your type of sub-interfaces (multi-point or p-t-p), which may depend on type of topology used: full-mesh or partial mesh and IP scheme.
So, I like their option of p-t-p subinterfaces as it gives you more flexibility of adding more PVCs later (as opose to tearing all down and re-building)
Closer look at picture, I can see now that R6 is directly connected to R4, which is spoke for R2, so likely you want to keep the R4-R6 on separate subnet form R4,R2,R5; separate subnet = PVC =point-to-point sub-interfaces.
Message was edited by: Martin